Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Characterization of EMF Exposure in Massive

MIMO Antenna Networks with Max-Min Fairness


Power Control
Maarouf Al Hajj∗ , Shanshan Wang∗ , Joe Wiart∗ ,
∗ Chaire C2M, LTCI, Télécom Paris, Palaiseau, France, maarouf.alhajj@telecom-paris.fr

Abstract—In this paper, we analyze the EMF exposure, in Many papers have addressed modeling the 5G network
terms of total received power, in the massive multiple-input using stochastic geometry however, power control was mostly
multiple-output (MIMO) networks. With the recent deployment missing since they considered the typical user. In [3] a
of 5G networks, the potential risks of electromagnetic field (EMF)
exposure are gaining increasing attention. However, most of millimeter-wave network has been shown to outperform the
the current research that focus on the mathematical modeling ultra-high frequency one for a dense 5G network. And in [4],
of 5G networks ignore downlink power control. Therefore, we 5G heterogeneous networks were modeled by PPP and K-
derive the framework of the average power received at nearest means clustering methods and their performance analyzed. As
mobile terminal MT under max-min fairness power control for EMF exposure studies of 5G networks, statistical studies
using stochastic geometry. The total received power consists of
three parts, useful signal, multi-user interference and inter-cell were performed to estimate more realistic power levels [5] for
interference. We propose a tight approximation on the power simple network architecture. And recently, some studies opted
control coefficient. Based on the proposed approximation, the to analyze the average EMF exposure in the cell for randomly
framework on total received power is then validated by Monte- distributed BSs following a Poisson point process (PPP) [6],
Carlo simulations. The results show that the average received [7].
power monotonically increases as the density of the base station
increases and the number of users increases. In this paper, we aim to characterize the power received
at the nearest MT0 to the serving BS in 5G massive MIMO
Index Terms—Cellular networks, massive MIMO, max-min networks. We consider a more realistic channel model with
power control, stochastic geometry, dosimetry.
max-min power control amongst all MTs. Then, a tight
approximation is proposed and validated by Monte-Carlo
I. I NTRODUCTION
simulations in order to obtain closed-form expression on the
Radio frequency (RF) electromagnetic field (EMF) compli- total received power at MT0 .
ance assessments are critical for manufacturers and operators
to prove that the transmitting equipment is compliant with II. S YSTEM M ODEL
safety regulations on human exposure and to determine exclu- In this section, we present the system model that we use
sion bounds around base stations (BS)s. In the far field, the to evaluate the total received power at M0 in a massive
considered EMF exposure metric is usually the power density MIMO network. Throughout this study, we assume the MRT
which is directly proportional to the received power [1]. precoding scheme at the transmitting BS.
Traditionally, RF EMF exposure assessments are conducted
assuming the maximum theoretical transmit power at the A. Massive MIMO Network Modeling
BS [2]. This will introduce an unnecessary bottleneck when We consider a downlink multi-cell, multi-user massive
deploying large, high-gain, massive multiple-input multiple- MIMO network. This network consists of BSs distributed
output (MIMO) antennas for 5G networks which concentrate following a homogeneous PPP ΦBS ⊂ R2 with density λBS .
the transmit power towards the served mobile terminal (MT). The MTs are modeled as another homogeneous PPP, ΦM T
The large massive MIMO allows the simultaneous transmis- with density λM T . We assume the zero-cell to be a disk
sion to multiple MTs in the cell in the same time-frequency centered around the BS and having a radius of R = √πλ1 ,
BS
resource block, through spacial multiplexing. To ensure fair which is the average size of the Voronoi cell of a homogeneous
distribution of resources, power control schemes exist to PPP. Since ΦBS is stationary and isotropic, the MTs will be
ensure sufficient resource allocation in the cell to fulfill certain distributed
 inside the cell also following a homogeneous PPP
quality criteria, either by allocating a minimal signal to inter- ΦlM T = Xkl , where Xkl is the k + 1-th nearest MT from
ference and noise ratio (SINR) targets or insuring equal SINR its serving BS in cell l. We condition ΦM T on having fixed
between MTs using max-min fairness power control. Massive number of points K in a single cell, representing the served
MIMO has been extensively studied throughout the years MTs in a single time-frequency block. The MTs in a single cell
with the main focus being the performance, and efficiency. l are therefore distributed following a binomial point process
However, the RF EMF exposure cannot be overlooked. (BPP) [8] ΦlM T with density λlM T = K/R. We denote by Rlk
TABLE I
TABLE OF VARIABLES AND SYMBOLS USED THROUGHOUT THE PAPER

Symbol Description
ΦBS , λBS , ΦlM T , λlM T Point process and density of the BSs and MTs respectively
R Radius of the 0-cell
K, M Number of simultaneously served MTs and number of antenna elements at each BS
BSl , M Tk BS in cell j and user k in the 0-cell
α Path-loss exponent
d Radius of the guard zone around the BSs
ρdl Downlink transmit power (mW)
g lk , β S
lk , hlk M × 1 channel, large scale, and small scale gain vectors between user k and its associated BS in cell l
ηlk Power control coefficient for MTlk
Rlk , Rk Distance between MTk and BSl , and association distance of user k in the 0-cell
(·)∗ , (·)H , | · | Complex conjugate and conjugate transpose of a matrix, respectively
El , El0 Average power received at a the typical MT and nearest MT to BSl respectively

the distance between the k-th MT to the l-th BS, and by Rk small-scale fadings, and sl is the transmitted signal from the
PK √
the association distance of MTk . Each BS is equipped with BS in cell l. It can be expressed as sl = k=1 ηlk alk qk ,
M antenna elements and each MT is equipped with a single with ηlk being the power control coefficient, and qk ∼
omnidirectional antenna. CN (0, 1) the transmit data symbols for the kth MT in cell
We define the M ×1 uplink random channel vector between l. Moreover, alk represents the vector of the linear precoder.
M Tk and BSl as the following Under MRT, the M × 1 precoding vector alk = [a1lk , ..., aM lk ]
T
p is
g lk = βlk hlk , (1) 1
alk = h i h∗lk , (4)
where βlk and hlk are the uncorrelated large-scale and small- E tr(Al AH l )
scale fading vectors respectively. We assume that the network
operates using a time-division duplex (TDD) scheme, to where Al is the M × K precoding matrix, the term
1
exploit channel reciprocity [9]. Thanks to these assumptions, E[tr(Al AH
= √1M is the normalization parameter insuring
l )]
the BS will be able to estimate the downlink channel matrix  
the constraint on the transmit power E ρdl sl (sl )H = ρdl .
using a pilot signal sent from the MT. We define the the fading Taking into account the previous assumptions, the total
vectors in the following. power received at MT0 in cell l, is determined based on the
1) Large-Scale Fading: We introduce a two-slope large-
results presented in [9] as follows
scale fading model representing the path-loss between BSl and
X X
MTk . In our scenario, it is essential to consider a two-region Pl0 =M ρdl ηl0 βl0 + ρdl βi0 + ρdl βl0 ηlk ,
large-scale fading coefficient to avoid the singularity for Rlk = i∈ΦBS \{l} k∈ΦM T \{X0 }
0. The large-scale fading coefficient is expressed by (5)
−α
βlk (Rlk ) = max (d, Rlk ) , (2) where the first term in (5) represents the received signal, the
+
where α is the path-loss exponent, and d ∈ R is the distance second term represents the inter-cell interference, and the last
of the guard zone around the BS. term represents the multi-user interference.
2) Small-Scale Fading: For a generic link between MTk
and BSl we assume that hlk ∼ CN (0, IM ) is an i.i.d. Rayleigh C. Max-Min Fairness Power Control
fading vector representing the small-scale fading of the NLoS Max-min power control seeks to maximize the worst SINR
link between an MT and a BS. in the cell by providing equal SINR for all the terminals. The
B. Downlink Transmission proposed model assumes no coordination between the BSs so
that the power control coefficient is computed solely based
We assume perfect CSI at the transmitter and no pilot on the multi-user channel inside the cell, and the coefficient is
contamination. The perfect CSI assumption produces a con- determined independently in each cell. This results in a simple
servative estimation of the downlink exposure. The signal optimization problem that produces a close-form solution for
received at the nearest MT, MT0 is given by this particular scenario as shown in [9]. We can write the power
√ X
control coefficient at the nearest MT to its BS ηl0 as
y 0 = ρdl gHl0 sl + w0 , (3)
l∈ΦBS
1 + ρdl βl0
where ρdl is the downlink transmit power, g l0 is the M × 1 ηl0 = P 1+ρdl βlk (6)
βl0
channel gain vector at MT0 taking into account large-scale and k∈ΦM T
βlk
This power control scheme introduces an important trade- and the multi-user interference part by
off in the power received at MT0 . Even though MT0 has !
the smallest large-scale fading, it will have fewer high-gain 1 + ρdl Ql0
IM = ρdl Ql0 −  , (13)
resources allocated to it, as per (5) K ρdl + Ῡ
We note that the power control coefficient ηlk is chosen
under the constraint where Ql0 = E [βl0 ] is the expectation of the large-scale fading
 allowing maximum power usage at each
BS, i.e. E ksl k2 = 1, ∀l, giving coefficient at MT0 in cell l. In the following, we derive the
X close form equations for each element in (9).
ηlk = 1, ∀l (7)
k∈ΦlM T
Lemma 1. The expectation of the large scale fading for the
nearest MT to the origin, Ql0 with path loss model in (2) is
From (7), we can rewrite (5) to be presented in (14) at the top of the next page. In which Γ(x)
is the gamma function, B(a, b) is the binomial function, and
X
Pl0 = M ρdl ηl0 βl0 + ρdl βi0 + ρdl βl0 (1 − ηl0 ) (8)
i∈ΦBS \{l} 2 F̃1 is the regularized hypergeometric function.

III. AVERAGE P OWER R ECEIVED Proof. In a point process consisting of K uniformly dis-
In this section, we derive the close-form expressions of the tributed points in a 2-dimensional ball of radius R centered at
average received power at the nearest MT in cell l, MT0 , using the origin b(0, d), the Euclidean distance Rlk from the origin to
an approximation of the power control coefficient. From this the kth nearest neighbor follows a generalized beta distribution
section onward, we ignore the noise factor w0 in the evaluation of the first kind with probability density function (PDF) [8]
of the total power received.
Without loss of generality, we assume that the typical BS 2 B(k + 1/2, K − k + 1)
fRlk (r) =
is located at the origin. From (5), knowing that the precoding d B(K − k + 1, k)
gain is independent of the power control coefficient and the  r 2
! (16)
large-scale coefficient, we can express the average instanta- ×β ; k + 1/2, K − k + 1 ,
d
neous power received at the nearest MT to its associated BS
as where β(x; a, b) = (1/B(a, b))xa−1 (1 − x)b−1 is the gen-
n
El0 = S + IM + II , (9) eralized beta function. The nth moment of Rlk , E [Rlk ] for
d < rk < R can be determined as
where S = ρdl M E [ηl0 βl0 ] is the average useful signal power,
IM = ρdl E [(1 − ηl0 )βl0 ] is the average multi-user interfer- 2
Z R   r 2 K−1
n n+2
hP i E [Rlk ] = 3 r 1− dr (17)
ence power, and II = E β
i∈ΦBS \{l} l0 is the average R B(K, 1) d R
inter-cell interference power. The sum in the denominator Γ n+3

Γ(K)Rn+3
of the power control coefficient (6) makes the derivation of = 2  − dn+3
2R2 Γ K + n+3 2
the expectations intractable. For this reason, we propose an ! (18)
n + 3 n + 5 d2

approximation for this parameter in the following proposition.
2 F̃1 1 − K, ; ; 2 ,
Proposition 1. For MT0 in cell l, the power control coefficient 2 2 R
ηl0 can be approximated, for large number of served users K,
for 0 < rk < d, E [βk ] can be determined likewise by
using the weak law of large numbers [10] as in
replacing rn with dn inside the integral. The summation of
1 + ρdl βl0 the expression in the two ranges gives (14) for n = −α.
ηl0 = ! (10)
P −1
βl0 ρdl K + βlj Lemma 2. The average of the inverse of the large-scale fading
j∈ΦM T
coefficient from a typical MT, Ῡ, can be formulated as follows,
1 + ρdl βl0
  (11) αdα+2 + 2Rα+2
Kβl0 ρdl + Ῡ Ῡ = (19)
h i (α + 2)R2
−1
where Ῡ = E βlj is the expectation of the inverse of the
large-scale fading coefficient from a typical MT in the cell. We Proof. We denote by FRk (x) = P (k ∈ Bd ) the probability
prove the validity of this approximation for our model with density of Rl which is the probability that the user k falls
numerical simulations presented in Figure 2b. For minimal inside the ball Bd = {b(0, x) : x ≤ R}. Its PDF can be
error, we assume K > 10 throughout this paper. determined as follows,

From proposition 1, we can express the useful signal part d Pr [k ∈ b(0, x)] 2x
fRk (x) = = 2. (20)
of the total received power by dx R
1 + ρdl Ql0 Using the path loss model in 2 the nth moment of the distance
S = M ρdl , (12) between the origin and a point distributed according to a BPP
K ρdl + Ῡ
K !
dn−1 d2 1 d2
 
1
d2 (2K − 1) + R2 − R2 2 F1

Ql0 = 1− − , 1 − K; ; 2
4K 2 − 1 R2 2 2 R
n+3
 ! (14)
Γ 2 Γ(K)R n+3 
n+3 n+5 d 2
n+3
+ − d 2 F̃1 1 − K, ; ; .
2R2 Γ K + n+3 R2

2
2 2
 
ρ2dl (M − 1) 2R−α
 
ρdl (M − 1)
El0 = Ql0   α+2 α+2
 +  α+2 α+2
 + ρdl πλBS R2 d−α + (15)
K ρdl + αd (α+2)R+2R
K ρdl + αd (α+2)R +2R α−2
2 2

in a ball b(0, R) centered at the origin according to the model


in (2) is
Z R
E [Rln ] = xn fRk (x) dx (21)
0
n dn+2 + 2Rn+2
= , (22)
(n + 2) R2
replacing n with α completes the proof.
Lemma 3. The expected value of the inter-cell interference is
2R−α
 
2 −α
II = πλBS R d + − Ql0 (23)
α−2
Proof. Following from the stationarity of ΦBS , this expression
can be simply determined using Campbell’s theorem
" #
X
II = E βi0 − βl0 (24)
i∈ΦBS Fig. 1. Total power received as function of the BS density for different
! number of simultaneously served users in the cell, K. Dotted line: Monte-
Z R Z ∞
Carlo simulation. Markers: Result from (15)
= 2πλBS d−α r dr + r1−α dr − Ql0 (25)
0 R

Solving the integrations gives (23) and completes the proof. i.e. increasing MT density. This is due to the average distance
between the BS and MT0 decreasing when the MT density
increases. However, in Fig. 2b we notice that this effect, for
The expression of the average total received power, El0 , is
the signal power, is overshadowed by the effect of resource di-
presented in (15) on the top of the previous page.
vision between the MTs. In addition to that, the big difference
IV. N UMERICAL R ESULTS in power between the multi-user interference and useful signal
In this section, we present the results of our model alongside at MT0 can be attributed to the power control itself, where the
their verification using Monte-Carlo simulations. This takes nearest MT will get the smallest fraction of transmit power,
into account the assumptions of the circular distribution of reducing the amplified useful signal, while receiving higher
MTs. The simulation parameters, unless precised otherwise, values of interference for its close proximity to the BS.
are chosen as M = 128, ρdl = 0 dBm, and α = 4. In all V. C ONCLUSION
figures, dotted lines represent the Monte-Carlo simulation and
In this paper, we have we derived the expression of the
the markers represent the values of the developed framework.
average power received at nearest mobile terminal to its
We plot the received powers for different values of simul-
serving BS, MT0 , under max-min fairness power control using
taneously served users K as function of the BS density λBS .
stochastic geometry. We determined the close-form expres-
From Fig. 2a, 2b, and 2c we notice that the received signal
sions of the useful received signal, the multi-user interference
power and the inter-cell interference power are negligible
and inter-cell interference. We proposed a tight approximation
compared to the interference levels, and the total received
on the power control coefficient, and based on it, the frame-
power in Fig. 1 is indistinguishable from the multi-user
work on total received power is then validated by Monte-Carlo
interference power. However, the received signal power is
simulations. The results show that the average received power
increasing exponentially and at a much higher rate than the
monotonically increases as the density of the base station
interference power which highlights the effect of network
increases and the number of users increases.
densification on the total exposure.
We can also notice from Fig. 1 the increase in the average
received power when increasing the number of served MTs,
(a) (b) (c)
Fig. 2. Multi-user interference (a), useful signal (b), and inter-cell interference (c) powers as function of the BS density for different number of simultaneously
served users in the cell, K. Dotted line: Monte-Carlo simulation. Markers: Result from (13).

VI. R EFERENCES Access, vol. 5, pp. 19 711–19 719, 2017, ISSN: 2169-
[1] R. Vallauri, G. Bertin, B. Piovano, and P. Gianola, 3536. DOI: 10.1109/ACCESS.2017.2753459.
“Electromagnetic field zones around an antenna for [6] Q. Gontier, L. Petrillo, F. Rottenberg, F. Horlin, J.
human exposure assessment: Evaluation of the human Wiart, C. Oestges, and P. De Doncker, “A stochastic
exposure to emfs.,” IEEE Antennas and Propagation geometry approach to emf exposure modeling,” IEEE
Magazine, vol. 57, no. 5, pp. 53–63, 2015. DOI: 10 . Access, vol. 9, pp. 91 777–91 787, 2021. DOI: 10.1109/
1109/MAP.2015.2474127. ACCESS.2021.3091804.
[2] IEC 62232:2017, “Determination of rf field strength, [7] M. Al Hajj, S. Wang, L. Thanh Tu, S. Azzi, and J.
power density and sar in the vicinity of radiocommuni- Wiart, “A Statistical Estimation of 5G Massive MIMO
cation base stations for the purpose of evaluating human Networks’ Exposure Using Stochastic Geometry in
exposure,” International Electrotechnical Commission, mmWave Bands,” Applied Sciences, vol. 10, no. 23,
Standard, Oct. 2017. p. 8753, 23 Jan. 2020. DOI: 10 . 3390 / app10238753.
[Online]. Available: https : / / www . mdpi . com / 2076 -
[3] T. Bai and R. W. Heath, “Coverage and Rate Analysis
3417/10/23/8753 (visited on 01/25/2021).
for Millimeter-Wave Cellular Networks,” IEEE Trans-
[8] S. Srinivasa and M. Haenggi, “Distance Distributions in
actions on Wireless Communications, vol. 14, no. 2,
Finite Uniformly Random Networks: Theory and Appli-
pp. 1100–1114, Feb. 2015, ISSN: 1558-2248. DOI: 10.
cations,” IEEE Transactions on Vehicular Technology,
1109/TWC.2014.2364267.
vol. 59, no. 2, pp. 940–949, Feb. 2010, ISSN: 1939-
[4] T. Maksymyuk, M. Brych, and V. Pelishok, “Stochastic
9359. DOI: 10.1109/TVT.2009.2035044.
geometry models for 5G heterogeneous mobile net-
[9] T. L. Marzetta, Fundamentals of Massive MIMO, en.
works,” The Smart Computing Review, pp. 89–101,
2016, ISBN: 978-1-316-79989-5 978-1-107-17557-0.
2015. DOI: 10 . 6029 / smartcr. 2015 . 02 . 002. [Online].
[10] M. Loeve, Probability Theory I, 4th ed., ser. Graduate
Available: https://app.dimensions.ai/details/publication/
Texts in Mathematics, Graduate Texts MathematicsLo-
pub.1073602178.
eve,M.:Probability Theory. New York: Springer-Verlag,
[5] B. Thors, A. Furuskär, D. Colombi, and C. Törnevik,
1977, ISBN: 978-0-387-90210-4. DOI: 10.1007/978-1-
“Time-Averaged Realistic Maximum Power Levels for
4684-9464-8. [Online]. Available: https://www.springer.
the Assessment of Radio Frequency Exposure for 5G
com/gp/book/9780387902104 (visited on 09/30/2021).
Radio Base Stations Using Massive MIMO,” IEEE

You might also like